Transcript kiferComp_348761_ppt14

Chapter 14
Object Databases
1
What’s in This Module?
•
•
•
•
Motivation
Conceptual model
SQL:1999/2003 object extensions
ODMG
– ODL – data definition language
– OQL – query language
• CORBA
2
Problems with Flat Relations
Consider a relation
Person(SSN, Name, PhoneN, Child)
with:
• FD: SSN  Name
• Any person (identified by SSN) can have several
phone numbers and children
• Children and phones of a person are not related to
each other except through that person
3
An Instance of Person
SSN
Name
PhoneN
Child
111-22-3333
111-22-3333
Joe Public
Joe Public
Joe Public
Joe Public
516-123-4567
516-345-6789
516-123-4567
516-345-6789
222-33-4444
222-33-4444
333-44-5555
333-44-5555
222-33-4444
222-33-4444
222-33-4444
222-33-4444
Bob Public
Bob Public
Bob Public
Bob Public
212-987-6543
212-987-1111
212-987-6543
212-987-1111
444-55-6666
555-66-7777
555-66-7777
444-55-6666
111-22-3333
111-22-3333
4
Dependencies in Person
Join dependency (JD):
Person = (SSN,Name,PhoneN)
(SSN,Name,Child)
Functional dependency (FD):
SSN  Name
5
Redundancies in Person
• Due to the JD:
Every PhoneN is listed with every Child SSN
Hence Joe Public is twice associated with 222-33-4444
and with 516-123-4567
Similarly for Bob Public and other phones/children
• Due to the FD:
Joe Public is associated with the SSN 111-22-3333 four
times (for each of Joe’s child and phone)!
Similarly for Bob Public
6
Dealing with Redundancies
• What to do? Normalize!
– Split Person according to the JD
– Then each resulting relation using the FD
– Obtain four relations (two are identical)
7
Normalization removes redundancy:
Phone
Person1
SSN
PhoneN
111-22-3333
516-345-6789
SSN
Name
111-22-3333
Joe Public
111-22-3333
516-123-4567
222-33-4444
Bob Public
222-33-4444
212-987-6543
222-33-4444
212-135-7924
SSN
Child
111-22-3333
222-33-4444
111-22-3333
333-44-5555
222-33-4444
444-55-6666
222-33-4444
555-66-7777
ChildOf
8
But querying is still cumbersome:
Get the phone numbers of Joe’s grandchildren.
Against the original relation: three cumbersome joins
SELECT G.PhoneN
FROM
Person P, Person C, Person G
WHERE P.Name = ‘Joe Public’ AND
P.Child = C.SSN AND C.Child = G.SSN
Against the decomposed relations is even worse: four joins
SELECT N.PhoneN
FROM Person1 P, ChildOf C, ChildOf G, Phone N
WHERE P.Name = ‘Joe Public’ AND P.SSN = C.SSN AND
C.Child = G.SSN AND G.Child = N.SSN
9
Objects Allow Simpler Design
Schema:
Person(SSN: String,
Name: String,
PhoneN: {String},
Child: {SSN} )
Set data types
No need to decompose in order to eliminate redundancy:
the set data type takes care of this.
Object 1:
( 111-22-3333,
“Joe Public”,
{516-345-6789, 516-123-4567},
{222-33-4444, 333-44-5555}
)
Object 2:
( 222-33-4444,
“Bob Public”,
{212-987-6543, 212-135-7924},
{444-55-6666, 555-66-7777}
)
10
Objects Allow Simpler Queries
Schema (slightly changed):
Person(SSN: String,
Name: String,
PhoneN: {String},
Child: {Person})
Set of persons
- Because the type of Child is the set of Person-objects, it makes sense
to continue querying the object attributes in a path expression
Object-based query:
SELECT P.Child.Child.PhoneN
FROM Person P
WHERE P.Name = ‘Joe Public’
Path expression
- Much more natural!
11
ISA (or Class) Hierarchy
Person(SSN, Name)
Student(SSN,Major)
Query: Get the names of all computer science majors
Relational formulation:
SELECT P.Name
FROM
Person P, Student S
WHERE P.SSN = S.SSN and S.Major = ‘CS’
Object-based formulation:
SELECT S.Name
FROM Student S
WHERE S.Major = ‘CS’
Student-objects are also Person-objects, so they inherit the attribute Name
12
Object Methods in Queries
• Objects can have associated operations
(methods), which can be used in queries.
For instance, the method frameRange(from,
to) might be a method in class Movie. Then
the following query makes sense:
SELECT M.frameRange(20000, 50000)
FROM Movie M
WHERE M.Name = ‘The Simpsons’
13
The “Impedance” Mismatch
• One cannot write a complete application in SQL, so SQL
statements are embedded in a host language, like C or
Java.
• SQL: Set-oriented, works with relations, uses high-level
operations over them.
• Host language: Record-oriented, does not understand
relations and high-level operations on them.
• SQL: Declarative.
• Host language: Procedural.
• Embedding SQL in a host language involves ugly adaptors
(cursors/iterators) – a direct consequence of the above
mismatch of properties between SQL and the host
languages. It was dubbed “impedance” mismatch.
14
Can the Impedance Mismatch be Bridged?
• This was the original idea behind object databases:
Use an object-oriented language as a data manipulation language.
Since data is stored in objects and the language manipulates
objects, there will be no mismatch!
• Problems:
• Object-oriented languages are procedural – the advantages of a
high-level query language, such s SQL, are lost
• C++, Java, Smalltalk, etc., all have significantly different object
modeling capabilities. Which ones should the database use? Can a
Java application access data objects created by a C++ application?
• Instead of one query language we end up with a bunch! (one for
C++, one for Java, etc.)
15
Is Impedance Mismatch Really a
Problem?
• The jury is out
• Two main approaches/standards:
– ODMG (Object Database Management Group):
Impedance mismatch is worse than the ozone
hole!
– SQL:1999/2003:
Couldn’t care less – SQL rules!
• We will discuss both approaches.
16
Object Databases vs. Relational Databases
•
•
•
•
Relational: set of relations; relation = set of tuples
Object: set of classes; class = set of objects
Relational: tuple components are primitive (int, string)
Object: object components can be complex types (sets, tuples,
other objects)
• Unique features of object databases:
– Inheritance hierarchy
– Object methods
– In some systems (ODMG), the host language and the data manipulation
language are the same
17
The Conceptual Object Data Model
(CODM)
• Plays the same role as the relational data
model
• Provides a common view of the different
approaches (ODMG, SQL:1999/2003)
• Close to the ODMG model, but is not
burdened with confusing low-level details
18
Object Id (Oid)
• Every object has a unique Id: different
objects have different Ids
• Immutable: does not change as the object
changes
• Different from primary key!
– Like a key, identifies an object uniquely
– But key values can change – oids cannot
19
Objects and Values
• An object is a pair: (oid, value)
• Example: A Joe Public’s object
(#32, [ SSN: 111-22-3333,
Name: “Joe Public”,
PhoneN: {“516-123-4567”, “516-345-6789”},
Child: {#445, #73} ] )
20
Complex Values
• A value can be of one of the following forms:
– Primitive value: an integer (eg, 7), a string (“John”), a
float (eg, 23.45), a Boolean (eg, false)
– Reference value: An oid of an object, e.g., #445
– Tuple value: [A1: v1, …, An: vn]
– A1, …, An – distinct attribute names
– v1, …, vn – values
– Set value: {v1, …, vn}
– v1, …, vn
– values
• Complex value: reference, tuple, or set.
• Example: previous slide
21
Classes
• Class: set of semantically similar objects (eg,
people, students, cars, motorcycles)
• A class has:
– Type: describes common structure of all objects in the
class (semantically similar objects are also structurally
similar)
– Method signatures: declarations of the operations that
can be applied to all objects in the class.
– Extent: the set of all objects in the class
• Classes are organized in a class hierarchy
– The extent of a class contains the extent of any of its
subclasses
22
Complex Types: Intuition
• Data (relational or object) must be properly
structured
• Complex data (objects) – complex types
Object: (#32, [ SSN:
111-22-3333,
Name: “Joe Public”,
PhoneN: {“516-123-4567”, “516-345-6789”},
Child:
{#445, #73} ] )
Its type: [SSN:
String,
Name: String,
PhoneN: {String},
Child: {Person} ]
23
Complex Types: Definition
• A type is one of the following:
– Basic types: String, Float, Integer, etc.
– Reference types: user defined class names, eg, Person,
Automobile
– Tuple types: [A1: T1, …, An: Tn]
– A1, …, An – distinct attribute names
– T1, …, Tn – types
• Eg, [SSN: String, Child: {Person}]
– Set types: {T}, where T is a type
• Eg, {String}, {Person}
• Complex type: reference, tuple, set
24
Subtypes: Intuition
• A subtype has “more structure” than its supertype.
• Example: Student is a subtype of Person
Person: [SSN: String, Name: String,
Address: [StNum: Integer, StName: String]]
Student: [SSN: String, Name: String,
Address: [StNum: Integer, StName: String, Rm: Integer],
Majors: {String},
Enrolled: {Course} ]
25
Subtypes: Definition
• T is a subtype of T’ iff T  T’ and
– Reference types:
T, T’ are reference types and T is a subclass T’
– Tuple types:
T = [A1: T1, …, An: Tn, An+1: Tn+1, …, Am: Tm],
T’ = [A1: T1’, …, An: Tn’ ]
are tuple types and for each i=1,…,n, either Ti = Ti’ or Ti is a subtype of Ti’
– Set types:
T = {T0} and T ’ = {T0’ } are set types and T0 is a subtype of T0’
26
Domain of a Type
• domain(T) is the set of all objects that conform
to type T. Namely:
– domain(Integer) = set of all integers,
domain(String) = set of all strings, etc.
– domain(T), where T is reference type is the extent of
T, ie, oids of all objects in class T
– domain([A1: T1, …, An: Tn]) is the set of all tuple values
of the form [A1: v1, …, An: vn], where each vi domain(Ti)
– domain({T}) is the set of all finite sets of the form
{w1 , …, wm}, where each wi domain(T )
27
Database Schema
• For each class includes:
– Type
– Method signatures. E.g., the following signature could
be in class Course:
Boolean enroll(Student)
• The subclass relationship
• The integrity constraints (keys, foreign keys, etc.)
28
Database Instance
• Set of extents for each class in the schema
• Each object in the extent of a class must
have the type of that class, i.e., it must
belong to the domain of the type
• Each object in the database must have
unique oid
• The extents must satisfy the constraints of
the database schema
29
Object-Relational Data Model
• A straightforward subset of CODM: only tuple
types at the top level
• More precisely:
• Set of classes, where each class has a tuple type (the types of
the tuple component can be anything)
• Each tuple is an object of the form (oid, tuple-value)
• Pure relational data model:
• Each class (relation) has a tuple type, but
• The types of tuple components must be primitive
• Oids are not explicitly part of the model – tuples are pure
values
30
Objects in SQL:1999/2003
•
•
•
•
Object-relational extension of SQL-92
Includes the legacy relational model
SQL:1999/2003 database = a finite set of relations
relation = a set of tuples (extends legacy relations)
OR
•
•
•
•
a set of objects (completely new)
object = (oid, tuple-value)
tuple = tuple-value
tuple-value = [Attr1: v1, …, Attrn: vn]
multiset-value = {v1, …, vn }
31
SQL:1999 Tuple Values
• Tuple value: [Attr1: v1, …, Attrn: vn]
– Attri are all distinct attributes
– Each vi is one of these:
– Primitive value: a constant of type CHAR(…), INTEGER,
FLOAT, etc.
– Reference value: an object Id
– Another tuple value
– A collection value
MULTISET introduced in SQL:2003.
ARRAY– a fixed size array
32
Row Types
• The same as the original (legacy) relational tuple type.
However:
– Row types can now be the types of the individual attributes in
a tuple
– In the legacy relational model, tuples could occur only as toplevel types
CREATE TABLE PERSON (
Name CHAR(20),
Address ROW(Number INTEGER, Street CHAR(20), ZIP CHAR(5))
)
33
Row Types (Contd.)
• Use path expressions to refer to the components of row types:
SELECT P.Name
FROM PERSON P
WHERE P.Address.ZIP = ‘11794’
• Update operations:
INSERT INTO PERSON(Name, Address)
VALUES (‘John Doe’, ROW(666, ‘Hollow Rd.’, ‘66666’))
UPDATE PERSON
SET Address.ZIP = ‘66666’
WHERE Address.ZIP = ‘55555’
UPDATE PERSON
SET Address = ROW(21, ‘Main St’, ‘12345’)
WHERE
Address = ROW(123, ‘Maple Dr.’, ‘54321’) AND Name = ‘J. Public’
34
User Defined Types (UDT)
• UDTs allow specification of complex objects/tupes,
methods, and their implementation
• Like ROW types, UDTs can be types of individual
attributes in tuples
• UDTs can be much more complex than ROW types
(even disregarding the methods): the components of
UDTs do not need to be elementary types
35
A UDT Example
CREATE TYPE PersonType AS (
Name CHAR(20),
Address ROW(Number INTEGER, Street CHAR(20), ZIP CHAR(5))
);
CREATE TYPE StudentType UNDER PersonType AS (
Id INTEGER,
Status CHAR(2)
)
METHOD award_degree() RETURNS BOOLEAN;
CREATE METHOD award_degree() FOR StudentType
LANGUAGE C
EXTERNAL NAME ‘file:/home/admin/award_degree’;
File that holds the binary code
36
Using UDTs in CREATE TABLE
• As an attribute type:
CREATE TABLE TRANSCRIPT (
Student StudentType,
CrsCode CHAR(6),
Semester CHAR(6),
Grade CHAR(1)
)
A previously defined UDT
• As a table type:
CREATE TABLE STUDENT OF StudentType;
Such a table is called typed table.
37
Objects
• Only typed tables contain objects (ie, tuples with oids)
• Compare:
CREATE TABLE STUDENT OF StudentType;
and
CREATE TABLE STUDENT1 (
Name CHAR(20),
Address ROW(Number INTEGER, Street CHAR(20), ZIP CHAR(5)),
Id
INTEGER,
Status CHAR(2)
)
• Both contain tuples of exactly the same structure
• Only the tuples in STUDENT – not STUDENT1 – have oids
• Will see later how to reference objects, create them, etc.
38
Querying UDTs
• Nothing special – just use path expressions
SELECT T.Student.Name, T.Grade
FROM
TRANSCRIPT T
WHERE T.Student.Address.Street = ‘Main St.’
Note: T.Student has the type StudentType. The attribute Name is
not declared explicitly in StudentType, but is inherited from
PersonType.
39
Updating User-Defined Types
• Inserting a record into TRANSCRIPT:
INSERT INTO TRANSCRIPT(Student,Course,Semester,Grade)
VALUES (????, ‘CS308’, ‘2000’, ‘A’)
The type of the Student attribute is StudentType. How does one
insert a value of this type (in place of ????)?
Further complication: the UDT StudentType is encapsulated,
ie, it is accessible only through public methods, which we
did not define
Do it through the observer and mutator methods provided
by the DBMS automatically
40
Observer Methods
• For each attribute A of type T in a UDT, an SQL:1999 DBMS is supposed to
supply an observer method, A: ( )  T, which returns the value of A (the notation
“( )” means that the method takes no arguments)
• Observer methods for StudentType:
• Id: ( )  INTEGER
• Name: ( )  CHAR(20)
• Status: ( )  CHAR(2)
• Address: ( )  ROW(INTEGER, CHAR(20), CHAR(5))
• For example, in
SELECT T.Student.Name, T.Grade
FROM
TRANSCRIPT T
WHERE T.Student.Address.Street = ‘Main St.’
Name and Address are observer methods, since T.Student is of type StudentType
Note: Grade is not an observer, because TRANSCRIPT is not part of a UDT,
but this is a conceptual distinction – syntactically there is no difference
41
Mutator Methods
• An SQL DBMS is supposed to supply, for each
attribute A of type T in a UDT U, a mutator method
A: T  U
For any object o of type U, it takes a value t of type T
and replaces the old value of o.A with t; it returns the
new value of the object. Thus, o.A(t) is an object of
type U
• Mutators for StudentType:
• Id: INTEGER  StudentType
• Name: CHAR(20)  StudentType
• Address: ROW(INTEGER, CHAR(20), CHAR(5))  StudentType
42
Example: Inserting a UDT Value
INSERT INTO TRANSCRIPT(Student,Course,Semester,Grade)
VALUES (
NEW StudentType( ).Id(111111111).Status(‘G5’).Name(‘Joe Public’)
.Address(ROW(123,’Main St.’, ‘54321’)) ,
‘CS532’,
‘S2002’,
‘A’
)
Add a value
for Id
Create a blank
StudentType object
Add a value for the
Address attribute
Add a value
for Status
‘CS532’, ‘S2002’, ‘A’ are primitive values for the attributes Course, Semester, Grade
43
Example: Changing a UDT Value
UPDATE TRANSCRIPT
SET Student = Student.Address(ROW(21,’Maple St.’,’12345’)).Name(‘John Smith’),
Grade = ‘B’
Change Name
Change Address
WHERE Student.Id = 111111111 AND CrsCode = ‘CS532’ AND Semester = ‘S2002’
• Mutators are used to change the values of the attributes Address
and Name
44
Referencing Objects
• Consider again
CREATE TABLE TRANSCRIPT (
Student StudentType,
CrsCode CHAR(6),
Semester CHAR(6),
Grade CHAR(1)
)
• Problem: TRANSCRIPT records for the same student refer to distinct
values of type StudentType (even though the contents of these
values may be the same) – a maintenance/consistency problem
• Solution: use self-referencing column (next slide)
– Bad design, which distinguishes objects from their references
– Not truly object-oriented
45
Self-Referencing Column
• Every typed table has a self-referencing column
– Normally invisible
– Contains explicit object Id for each tuple in the table
– Can be given an explicit name – the only way to enable
referencing of objects
CREATE TABLE STUDENT2 OF StudentType
REF IS stud_oid;
Self-referencing column
Self-referencing columns can be used in queries just like regular columns
Their values cannot be changed, however
46
Reference Types and Self-Referencing Columns
• To reference objects, use self-referencing columns + reference
types: REF(some-UDT)
CREATE TABLE TRANSCRIPT1 (
Student REF(StudentType) SCOPE STUDENT2,
CrsCode CHAR(6),
Semester CHAR(6),
Grade CHAR(1)
)
Reference type
Typed table where the
values are drawn from
• Two issues:
• How does one query the attributes of a reference type
• How does one provide values for the attributes of type REF(…)
– Remember: you can’t manufacture these values out of thin air – they are oids!
47
Querying Reference Types
• Recall:
Student REF(StudentType) SCOPE STUDENT2
in
TRANSCRIPT1.
How does one access, for example, student names?
• SQL:1999 has the same misfeature as C/C++ has (and which Java and
OQL do not have): it distinguishes between objects and references to
objects. To pass through a boundary of REF(…) use “” instead of “.”
SELECT T.StudentName, T.Grade
FROM TRANSCRIPT1 T
WHERE
T.StudentAddress.Street = “Main St.”
Not crossing REF(…)
boundary, use “.”
Crossing REF(…)
boundary, use 
48
Inserting REF Values
• How does one give values to REF attributes, like Student in
TRANSCRIPT1?
• Use explicit self-referencing columns, like stud_oid in STUDENT2
• Example: Creating a TRANSCRIPT1 record whose Student attribute has
an object reference to an object in STUDENT2:
INSERT INTO TRANSCRIPT1(Student,Course,Semester,Grade)
SELECT S.stud_oid, ‘HIS666’, ‘F1462’, ‘D’
FROM STUDENT2 S
WHERE S.Id = ‘111111111’
Explicit self-referential
column of STUDENT2
49
Collection Data Types
• Set (multiset) data type was added in SQL:2003.
CREATE TYPE StudentType UNDER PersonType AS (
Id INTEGER,
Status CHAR(2),
Enrolled REF(CourseType) MULTISET
)
A bunch of references to objects of
type CourseType
50
Querying Collection Types
• For each student, list the Id, address, and the
courses in which the student is enrolled
(assume STUDENT is a table of type
StudentType):
SELECT S.Id, S.Address, C.Name
Convert multiset
to table
FROM
STUDENT S, COURSE C
WHERE C.CrsCode IN
( SELECT E  CrsCode
FROM UNNEST(S.Enrolled) E)
• Note: E is bound to tuples in a 1-column
table of object references
51
The ODMG Standard
•
•
•
•
•
•
ODMG 3.0 was released in 2000
Includes the data model (more or less)
ODL: The object definition language
OQL: The object query language
A transaction specification mechanism
Language bindings: How to access an
ODMG database from C++, Smalltalk, and
Java (expect C# to be added to the mix)
52
The Structure of an ODMG Application
53
Main Idea: Host Language = Data Language
• Objects in the host language are mapped directly to
database objects
• Some objects in the host program are persistent. Think of
them as “proxies” of the actual database objects. Changing
such objects (through an assignment to an instance variable
or with a method application) directly and transparently
affects the corresponding database object
• Accessing an object using its oid causes an “object fault”
similar to pagefaults in operating systems. This
transparently brings the object into the memory and the
program works with it as if it were a regular object
defined, for example, in the host Java program
54
Architecture of an ODMG DBMS
55
SQL Databases vs. ODMG
• In SQL: Host program accesses the database by
sending SQL queries to it (using JDBC, ODBC,
Embedded SQL, etc.)
• In ODMG: Host program works with database
objects directly
• ODMG has the facility to send OQL queries to the
database, but this is viewed as evil: brings back
the impedance mismatch
56
ODL: ODMG’s Object Definition Language
• Is rarely used, if at all!
– Relational databases: SQL is the only way to describe data to the DB
– ODMG databases: can do this directly in the host language
– Why bother to develop ODL then?
• Problem: Making database objects created by applications
written in different languages (C++, Java, Smalltalk)
interoperable
– Object modeling capabilities of C++, Java, Smalltalk are very different.
– How can a Java application access database objects created with C++?
• Hence: Need a reference data model, a common target to which
to map the language bindings of the different host languages
– ODMG says: Applications in language A can access objects created by
applications in language B if these objects map into a subset of ODL
supported by language A
57
ODMG Data Model
• Classes + inheritance hierarchy + types
• Two kinds of classes: “ODMG classes” and “ODMG
interfaces”, similarly to Java
– An ODMG interface:
• has no method code – only signatures
• does not have its own objects – only the objects that belong to the interface’s
ODMG subclasses
• cannot inherit from (be a subclass of) an ODMG class – only from another
ODMG interface (in fact, from multiple such interfaces)
– An ODMG class:
• can have methods with code, own objects
• can inherit from (be a subclass of) other ODMG classes or interfaces
– can have at most one immediate superclass (but multiple immediate superinterfaces)
58
ODMG Data Model (Cont.)
• Distinguishes between objects and pure
values (values are called literals)
• Both can have complex internal structure, but only
objects have oids
59
Example
interface PersonInterface: Object {
// Object is the ODMG topmost interface
attribute String Name;
attribute String SSN;
Integer Age();
}
class PERSON: PersonInterface
// inherits from ODMG interface
( extent PersonExt
// note: extents have names
keys SSN, (Name, PhoneN) ) : persistent;
{ attribute ADDRESS Address;
attribute Set<String> PhoneN;
attribute enum SexType {m,f} Sex;
attribute date DateOfBirth;
relationship PERSON Spouse;
// note: relationship vs. attribute
relationship Set<PERSON> Child;
void add_phone_number(in String phone); // method signature
}
struct ADDRESS { // a literal type (for pure values)
String StNumber;
String StName;
60
}
More on the ODMG Data Model
• Can specify keys (also foreign keys – later)
• Class extents have their own names – this is what
is used in queries
• As if relation instances had their own names, distinct from the
corresponding tables
• Distinguishes between relationships and attributes
• Attribute values are literals
• Relationship values are objects
• ODMG relationships have little to do with relationships in the
E-R model – do not confuse them!!
61
Example (contd.)
class STUDENT extends PERSON {
( extent StudentExt )
attribute Set<String> Major;
relationship Set<COURSE> Enrolled;
}
is a subclass of PERSON (both are classes,
unlike ADDRESS in the previous example)
• A class can have at most one immediate superclass
• No name overloading: a method with a given
name and signature cannot be inherited from more
than one place (a superclass or super-interface)
• STUDENT
62
Referential Integrity
class STUDENT extends PERSON {
( extent StudentExt )
attribute Set<String> Major;
relationship Set<COURSE> Enrolled;
}
class COURSE: Object {
( extent CourseExt )
attribute Integer CrsCode;
attribute String Department;
relationship Set<STUDENT> Enrollment;
}
•
•
•
Referential integrity: If JoePublic takes CS532, and CS532  JoePublic.Enrolled,
then deleting the object for CS532 will delete it from the set JoePublic.Enrolled
Still, the following is possible:
CS532  JoePublic.Enrolled but JoePublic  CS532.Enrollment
Question: Can the DBMS automatically maintain consistency between
JoePublic.Enrolled and CS532.Enrollment?
63
Referential Integrity (Contd.)
Solution:
class STUDENT extends PERSON {
( extent StudentExt )
attribute Set<String> Major;
relationship Set<COURSE> Enrolled
inverse COURSE::Enrollment;
}
class COURSE: Object {
( extent CourseExt )
attribute Integer CrsCode;
attribute String Department;
relationship Set<STUDENT> Enrollment
inverse STUDENT::Enrolled;
}
64
OQL: The ODMG Query Language
• Declarative
• SQL-like, but better
• Can be used in the interactive mode
• Very few vendors support interactive use
• Can be used as embedded language in a host
language
• This is how it is usually used
• OQL brings back the impedance mismatch
65
Example: Simple OQL Query
SELECT DISTINCT S.Address
FROM PersonExt S
WHERE S.Name = “Smith”
• Can hardly tell if this is OQL or SQL
• Note: Uses the name of the extent of class
PERSON, not the name of the class
66
Example: A Query with Method Invocation
• Method in the SELECT clause:
SELECT M.frameRange(100, 1000)
FROM MOVIE M
WHERE M.Name = “The Simpsons”
• Method with a side effect:
SELECT S.add_phone_number(“555-1212”)
FROM PersonExt S
WHERE S.SSN = “123-45-6789”
67
OQL Path Expressions
• Path expressions can be used with attributes:
SELECT DISTINCT S.Address.StName
FROM PersonExt S
WHERE S.Name = “Smith”
Attribute
• As well as with relationships:
SELECT DISTINCT S.Spouse.Name
FROM PersonExt S
WHERE S.Name = “Smith”
Relationship
68
Path Expressions (Contd.)
• Must be type consistent: the type of each prefix of a
path expression must be consistent with the
method/attribute/relationship that follows
• For instance, is S is bound to a PERSON object, then
S.Address.StName and S.Spouse.Name are type
consistent:
• PERSON objects have attribute Address and relationship Spouse
• S.Address is a literal of type ADDRESS; it has an attribute StName
• S.Spouse is an object of type PERSON; it has a attribute Name,
which is inherited from PersonInterface
69
Path Expressions (Contd.)
•
Is P.Child.Child.PhoneN type consistent (P is bound to
a PERSON objects)?
–
•
In some query languages, but not in OQL!
Issue: Is P.Child a single set-object or a set of objects?
1.
2.
•
If it is a set of PERSON objects, we can apply Child to each such
object and P.Child.Child makes sense (as a set of grandchild
PERSON objects)
If it is a single set-object of type Set<PERSON>, then
P.Child.Child does not make sense, because such objects do not
have the Child relationship
OQL uses the second option. Can we still get the phone
numbers of grandchildren? – Must flatten out the sets:
flatten(flatten(P.Child).Child).Phone
– A bad design decision. We will see in Chapter 17 that
XML query languages use option 1.
70
Nested Queries
• As in SQL, nested OQL queries can occur in
– The FROM clause, for virtual ranges of variables
– The WHERE clause, for complex query conditions
• In OQL nested subqueries can occur in SELECT, too!
• Do nested subqueries in SELECT make sense in SQL?
What does the next query do?
SELECT struct{ name: S.Name,
courses: (SELECT E
FROM S.Enrolled E
WHERE E.Department=“CS”)
}
FROM StudentExt S
71
Aggregation and Grouping
• The usual aggregate functions avg, sum, count, min, max
• In general, do not need the GROUP BY operator, because
we can use nested queries in the SELECT clause.
– For example: Find all students along with the number of
Computer Science courses each student is enrolled in
SELECT name : S.Name
count: count( SELECT E.CrsCode
FROM S.Enrolled E
WHERE E.Department = “CS” )
FROM StudentExt S
72
Aggregation and Grouping (Contd.)
• GROUP BY/HAVING exists, but does not increase the
expressive power (unlike SQL):
SELECT S.Name, count: count(E.CrsCode)
FROM StudentExt S, S.Enrolled E
WHERE E.Department = “CS”
GROUP BY S.SSN
Same effect, but the optimizer can use it as a hint.
73
GROUP BY as an Optimizer Hint
SELECT
name : S.Name
count: count(SELECT E.CrsCode
FROM S.Enrolled E
WHERE E.Department = “CS” )
FROM StudentExt S
The query optimizer would compute the
inner query for each sStudentExt, so
s.Enrolled will be computed for each s.
If enrollment information is stored
separately (not as part of the STUDENT
Object), then given s, index is likely to be
used to find the corresponding courses.
Can be expensive, if the index is not
clustered
SELECT S.Name, count: count(E.CrsCode)
FROM StudentExt S, S.Enrolled E
WHERE E.Department = “CS”
GROUP BY S.SSN
The query optimizer can recognize that
it needs to find all courses for each
student. It can then sort the enrollment
file on student oids (thereby grouping
courses around students) and then
compute the result in one scan of that
sorted file.
74
ODMG Language Bindings
• A set of interfaces and class definitions that allow host
programs to:
– Map host language classes to database classes in ODL
– Access objects in those database classes by direct manipulation of
the mapped host language objects
• Querying
– Some querying can be done by simply applying the methods
supplied with the database classes
– A more powerful method is to send OQL queries to the database
using a statement-level interface (which makes impedance
mismatch)
75
Java Bindings: Simple Example
public class STUDENT extends PERSON {
public DSet Major;
……….
}
• DSet class
Cant say “set of strings” –
a Java limitation
– part of ODMG Java binding, extends Java Set class
– defined because Java Set class cannot adequately replace ODL’s Set<…>
STUDENT X;
………
X.Major.add(“CS”);
………
add( ) is a method of class DSet (a modified Java’s method). If X is
bound to a persistent STUDENT object, the above Java statement will
change that object in the database
76
Language Bindings: Thorny Issues
• Host as a data manipulation language is a powerful idea, but:
– Some ODMG/ODL facilities do not exist in some or all host languages
– The result is the lack of syntactic and conceptual unity
• Some issues:
– Specification of persistence (which objects persist, ie, are automatically
stored in the database by the DBMS, and which are transient)
• First, a class must be declared persistence capable (differently in different languages)
• Second, to actually make an object of a persistence capable class persistent, different
facilities are used:
– In C++, a special form of new() is used
– In Java, the method makePersistent() (defined in the ODMG-Java interface Database) is used
– Representation of relationships
– Java binding does not support them; C++ and Smalltalk bindings do
– Representation of literals
– Java & Smalltalk bindings do not support them; C++ does
77
Java Bindings: Extended Example
• The OQLQuery class:
class OQLQuery {
public OQLQuery(String query); // the query constructor
public bind(Object parameter); // explained later
public Object execute();
// executes queries
… … several more methods … …
}
• Constructor: OQLQuery(“SELECT …”)
– Creates a query object
– The query string can have placeholders $1, $2, etc., like the `?’
placeholders in Dynamic SQL, JDBC, ODBC. (Why?)
78
Extended Example (Cont.)
• Courses taken exclusively by CS students in Spring 2002:
DSet students,courses;
String semester;
OQLQuery query1, query2;
query1 = new OQLQuery(“SELECT S FROM STUDENT S “
+ “WHERE \”CS\” IN S.Major”);
students = (DSet) query1.execute();
query2 = new OQLQuery(“SELECT T FROM COURSE T “
+ “WHERE T.Enrollment.subsetOf($1) “
+ “AND T.Semester = $2”);
semester = new String(“S2002”);
query2.bind(students); // bind $1 to the value of the variable students
query2.bind(semester); // bind $2 to the value of the variable semester
courses = (DSet) query2.execute();
79
Interface DCollection
• Allows queries (select) from collections of database
objects
• DSet inherits from DCollection, so, for example, the
methods of DCollection can be applied to variables
courses, students (previous slide)
public interface DCollection extends java.util.Collection {
public DCollection query(String condition);
public Object selectElement(String condition);
public Boolean existsElement(String condition);
public java.util.Iterator select(String condition);
}
80
Extended Example (Cont.)
• query( condition) – selects a subcollection of objects that
satisfy condition:
DSet seminars;
seminars = (DSet) courses.query(“this.Credits = 1”);
• select(condition) – like query( ), but creates an iterator; can
now scan the selected subcollection object-by-object:
java.util.Iterator seminarIter;
Course seminar;
seminarIter = (java.util.Iterator) courses.select(“this.Credits=1”);
while ( seminar=seminarIter.next( ) ) {
………
}
81
CORBA:
Common Object Request Broker Architecture
• Distributed environment for clients to access objects on various
servers
• Provides location transparency for distributed computational
resources
• Analogous to remote procedure call (RPC) and remote method
invocation in Java (RMI) in that all three can invoke remote
code.
• But CORBA is more general and defines many more protocols
(eg, for object persistence, querying, etc.). In fact, RMI is
implemented using CORBA in Java 2
82
Interface Description Language (IDL)
• Specifies interfaces only (ie, classes without
extents, attributes, etc.)
• No constraints or collection types
// File Library.idl
module Library {
interface myLibrary{
string searchByKeywords(in string keywords);
string searchByAuthorTitle(in string author, in string title);
}
}
83
Object Request Broker (ORB)
• Sits between clients and servers
• Identifies the actual server for each method call
and dispatches the call to that server
• Objects can be implemented in different
languages and reside on dissimilar
OSs/machines, so ORB converts the calls
according to the concrete language/OS/machine
conventions
84
ORB Server Side
• Library.idl  IDL Compiler  Library-stubs.c, Library-skeleton.c
 Method signatures to interface repository
• Server skeleton: Library-skeleton.c
• Requests come to the server in OS/language/machine independent way
• Server objects are implemented in some concrete language, deployed on a
concrete OS and machine
• Server skeleton maps OS/language/machine independent requests to calls
understood by the concrete implementation of the objects on the server
• Object adaptor: How does ORB know which server can handle
which method calls? – Object adaptor, a part of ORB
• When a server starts, it registers itself with the ORB object adaptor
• Tells which method calls in which interfaces it can handle. (Recall that method
signature for all interfaces are recorded in the interface repository).
• Implementation repository: remembers which server implements
which methods/interfaces (the object adaptor stores this info when
a server registers)
85
ORB Client Side
• Static invocation: used when the application knows
which exactly method/interface it needs to call to get
the needed service
• Dynamic invocation: an application might need to figure out
what method to call by querying the interface repository
• For instance, an application that searches community libraries, where
each library provides different methods for searching with different
capabilities. For instance, some might allow search by title/author,
while others by keywords. Method names, argument semantics, even
the number of arguments might be different in each case
86
Static Invocation
• Client stub: Library-stubs.c
– For static invocation only, when the method/interface to call is known
– Converts OS/language/machine specific client’s method call into the
OS/language/machine independent format in which the request is delivered
over the network
– This conversion is called marshalling of arguments
– Needed because client and server can be deployed on different
OS/machine/etc.
– Consider: 32-bit machines vs. 64 bit, little-endian vs. big endian, different
representation for data structures (eg, strings)
– Recall: the machine-independent request is unmarshalled on the server side
by the server skeleton
– Conversion is done transparently for the programmer – the programmer
simply links the stub with the client program
87
Dynamic Invocation
• Used when the exact method calls are not known
• Example: Library search service
– Several community libraries provide CORBA objects for searching their
book holdings
– New libraries can join (or be temporarily or permanently down)
– Each library has its own legacy system, which is wrapped in CORBA
objects. While the wrappers might follow the same conventions, the search
capabilities of different libraries might be different (eg, by keywords, by
wildcards, by title, by author, by a combination thereof)
– User fills out a Web form, unaware of what kind of search the different
libraries support
– The user-side search application should
• take advantage of newly joined libraries, even with different search
capabilities
• continue to function even if some library servers are down
88
Dynamic Invocation (Contd.)
• Example: IDL module with different search capabilities
module Library {
interface library1 {
string searchByKeywords(in string keywords);
string searchByAuthorTitle(in string author, in string title);
}
interface library2 {
void searchByTitle(in string title, out string result);
void searchByWildcard(in string wildcard, out string result);
}
}
The client application:
•
•
•
•
Examines the fields in the form filled out by the user
Examines the interface repository – next slide
Decides which methods it can call with which arguments
Constructs the actual call – next slide
89
Dynamic Invocation API
• Provides methods to query the interface repository
• Provides methods to construct machine-independent requests to
be passed along to the server by the ORB
• Once the application knows which method/interface to call with
which arguments, it constructs a request, which includes:
–
–
–
–
–
Object reference (which object to invoke)
Operation name (which method in which interface to call)
Argument descriptors (argument names, types, values)
Exception handling info
Additional “context” info, which is not part of the method arguments
• Note: The client stub is essentially a piece of code that uses the
dynamic invocation API to create the above requests. Thus:
• With static invocation, the stub is created automatically by the IDL compiler
• With dynamic invocation, the programmer has to manually write the code to
create and invoke the requests, because the requisite information is not
available at compile time
90
CORBA Architecture
91
Interoperability within CORBA
• ORB allows objects to talk to each other if they are
registered with that ORB; can objects registered with
different ORBs talk to each other?
• General inter-ORB protocol (GIOP): message format
for requesting services from objects that live under the
control of a different ORB
– Often implemented using TCP/IP
– Internet inter-ORB protocol (IIOP) specifies how GIOP
messages are encoded for delivery via TCP/IP
92
Inter-ORB Architecture
93
CORBA Services
• Rich infrastructure on top of basic CORBA
• Some services support database-like functions:
• Persistence services – how to store CORBA objects in a
database or some other data store
• Object query services – how to query persistent CORBA
objects
• Transaction services – how to make CORBA applications
atomic (either execute them to the end or undo all changes)
• Concurrency control services – how to request/release
locks. In this way, applications can implement transaction
isolation policies, such as two-phase commit
94
Persistent State Services (PSS)
• PSS – a standard way for data stores (eg, databases, file
systems) to define interfaces that can be used by CORBA
clients to manipulate the objects in that data store
• On the server:
• Objects are in storage homes (eg, classes)
• Storage homes are grouped in data stores (eg, databases)
• On the client:
• Persistent objects (from the data store) are represented using storage
object proxies
• Storage object proxies are organized into storage home proxies
• Clients manipulate storage object proxies directly, like
ODMG applications do
95
CORBA Persistent State Services
96
Object Query Services (OQS)
• OQS makes it possible to query persistent CORBA objects
• Supports SQL and OQL
• Does two things:
– Query evaluator: Takes a query (from the client) and translated it into the
query appropriate for the data store at hand (eg, a file system does not support
SQL, so the query evaluator might have quite some work to do)
– Query collection service: Processes the query result.
• Creates an object of type collection, which contain references to the objects in the
query result
• Provides an iterator object to let the application to process each object in the
result one by one
97
Object Query Services
98
Transaction and Concurrency Services
• Transactional services:
– Allow threads to become transactions. Provide
• begin()
• rollback()
• commit()
– Implement two-phase commit protocol to ensure atomicity
of distributed transactions
• Concurrency control services:
– Allow transactional threads to request and release locks
– Implement two-phase locking
– Only supports – does not enforce – isolation. Other, nontransactional CORBA applications can violate serializability
99